Deepseek V4 Pro requires 3.7 times lower FLOPs (compute) compared to the previous Deepseek version 3.2.

The key idea behind Deepseek V4's hybrid attention architecture is not to treat all past information as equally important.

Deepseek V4's sliding window attention keeps the most recent tokens, such as the last 128 words, completely uncompressed with full fidelity.

Deepseek V4's data center choreography breaks down data transfer into smaller sequential waves, overlapping computation and communication to eliminate network latency.

DeepSeek's AI models perform as efficiently as their American counterparts.

Deepseek is making a permanent 75% price cut on its flagship V4 Pro AI model, reducing prices to a quarter of their original level.